Page 28 - Read Online

Page 23 Qi et al. Intell Robot 2021;1(1):18-57 I http://dx.doi.org/10.20517/ir.2021.02

Global Model

4 Aggregation
Updated Model Coordinator Updated Model

1 Initialization
3 Submit Submit

3

Local Model A Local Model B Local Model C
2
Train Train Train

Dataset A Dataset B Dataset C
Participant Participant Participant

5 Update Update

Figure 1. An example of federated learning architecture: Client-Server Model.

Updated Model B 1 Initialization Updated Model C
Aggregation
Exchange 4
3
Local Model A
Trained Model A Trained Model A
2 Train
Dataset A
Participant A

Initialization Updated Model A Updated Model A Initialization
Aggregation Aggregation

Local Model B Local Model C
Trained Model B Trained Model C
Train Train
Dataset B Dataset C

Participant B Participant C

Figure 2. An example of federated learning architecture: Peer-to-Peer Model.

The termination conditions of the process can be designed by participants according to their needs. This
architecture further guarantees security since there is no centralized server. However, it requires more com-
munication resources and potentially increased computation for more messages.